Both vCenter and Veeam use the same exact API calls to create and remove snapshots, so there's got to be some other difference in these experiments. Perhaps you can demonstrate the same live to our support engineer, who can then review the debug logs to find the difference.
For example, depending on your backup mode, there are other operations that will delay the beginning of snapshot removal, like hot remove of the backed up disk, and the debug logs will reflect these operations. Regardless though, the actual snapshot removal should take the same time in both cases with very small snapshots, because this operation is managed by ESXi and is performed by the same exact ESXi function (Veeam has no control over it whatsoever).
But that is assuming clean experiment, of course. For example, if that test backup job continues on to processing other VMs while the snapshot is being removed, this will put additional load on the datastore, thus slowing down the snapshot removal. Anyway, it should be super easy to get to the bottom of this difference. There has not been a month in the past 12 years of Veeam existence without similar concerns, and every time we were able to demonstrate all doubting customers the consistency of snapshot removal performance regardless of how it is initiated